TradingAgents: A Multi‑Agent LLM Framework for Financial Trading
TradingAgents is an open‑source Python framework that splits the trading workflow into five specialized LLM agents, uses structured JSON communication, supports multiple model providers, and lets users quickly backtest or run live strategies with a single pip install.
Core problem addressed
Traditional quantitative trading relies on rule‑based strategies written by analysts, but markets are nonlinear and multi‑factor. Using a single large language model for research, position sizing, and risk control introduces hallucinations and bias because the roles conflict, degrading decision quality.
Architectural highlights: five agents with distinct responsibilities
TradingAgents decomposes the workflow into five dedicated agents that form a miniature investment‑bank research team:
Research Manager – coordinates research direction.
Analyst – performs deep data mining and produces technical indicators, sentiment scores, and recommendation rationale.
Trader – executes buy/sell decisions.
Risk Manager – controls drawdown.
Portfolio Manager – allocates final positions.
Each agent outputs a structured JSON message rather than free‑form text, allowing downstream agents to parse reasoning results precisely and avoid miscommunication. The framework is built on LangGraph, providing checkpoint‑resume capability and persistent decision logs for full traceability of the reasoning chain.
Version v0.2.4 adds support for domestic models such as DeepSeek, Qwen, and GLM, as well as Azure deployment, removing the need to stay within the OpenAI ecosystem.
Quick start: run a strategy in five minutes
Install the package with a single command: pip install trading-agents Configure an LLM API key (OpenAI, Anthropic, Google, etc.) and launch the framework: trading-agents run --config config.yaml The CLI includes backtesting and live‑trading modes. In config.yaml specify the stock universe, time window, and model parameters; the agent team then handles the rest.
Python developers can also call the API directly or follow the provided Jupyter Notebook that walks from data acquisition to backtest visualization.
Intended users
Quantitative researchers can prototype LLM‑enhanced multi‑factor strategies; AI developers can explore multi‑agent collaboration practices; fintech founders can build SaaS products on top of the framework.
Technical report
The accompanying technical report Trading‑R1 (arXiv:2509.11420) details the inter‑agent reasoning mechanism.
Repository
Public repository: https://github.com/TauricResearch/TradingAgents
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.